The spec says:
> ⬅️ Initialized Event
> This event indicates that the debug adapter is ready to accept
configuration requests (e.g. setBreakpoints, setExceptionBreakpoints).
>
> A debug adapter is expected to send this event when it is ready to
accept configuration requests (but not before the initialize request has
finished).
Previously in tests, `intercept_debug_sessions` was just spawning off a
background task to send the event after setting up the client, so the
event wasn't actually synchronized with the flow of messages in the way
the spec says it should be. This PR makes it so that the `FakeTransport`
injects the event right after a successful response to the initialize
request, and doesn't send it otherwise.
Release Notes:
- N/A
As reported [in
Discord](https://discord.com/channels/869392257814519848/1106226198494859355/1398470747227426948)
C projects with `"` as "brackets" that autoclose, may invoke panics when
edited at the end of the file.
With a single selection-caret (`ˇ`), at the end of the file,
```c
ifndef BAR_H
#define BAR_H
#include <stdbool.h>
int fn_branch(bool do_branch1, bool do_branch2);
#endif // BAR_H
#include"ˇ"
```
gets an LSP response from clangd
```jsonc
{
"filterText": "AGL/",
"insertText": "AGL/",
"insertTextFormat": 1,
"kind": 17,
"label": " AGL/",
"labelDetails": {},
"score": 0.78725427389144897,
"sortText": "40b67681AGL/",
"textEdit": {
"newText": "AGL/",
"range": { "end": { "character": 11, "line": 8 }, "start": { "character": 10, "line": 8 } }
}
}
```
which replaces `"` after the caret (character/column 11, 0-indexed).
This is reasonable, as regular follow-up (proposed in further
completions), is a suffix + a closing `"`:
<img width="842" height="259" alt="image"
src="https://github.com/user-attachments/assets/ea56f621-7008-4ce2-99ba-87344ddf33d2"
/>
Yet when Zed handles user input of `"`, it panics due to multiple
reasons:
* after applying any snippet text edit, Zed did a selection change:
5537987630/crates/editor/src/editor.rs (L9539-L9545)
which caused eventual autoclose region invalidation:
5537987630/crates/editor/src/editor.rs (L2970)
This covers all cases that insert the `include""` text.
* after applying any user input and "plain" text edit, Zed did not
invalidate any autoclose regions at all, relying on the "bracket" (which
includes `"`) autoclose logic to rule edge cases out
* bracket autoclose logic detects previous `"` and considers the new
user input as a valid closure, hence no autoclose region needed.
But there is an autoclose bracket data after the plaintext completion
insertion (`AGL/`) really, and it's not invalidated after `"` handling
* in addition to that, `Anchor::is_valid` method in `text` panicked, and
required `fn try_fragment_id_for_anchor` to handle "pointing at odd,
after the end of the file, offset" cases as `false`
A test reproducing the feedback and 2 fixes added: proper, autoclose
region invalidation call which required the invalidation logic tweaked a
bit, and "superficial", "do not apply bad selections that cause panics"
fix in the editor to be more robust
Release Notes:
- Fixed panic with completion ranges and autoclose regions interop
---------
Co-authored-by: Max Brunsfeld <maxbrunsfeld@gmail.com>
Follow-up of https://github.com/zed-industries/zed/pull/34208
Closes https://github.com/zed-industries/zed/issues/35185
Previous code assumed that extensions' language server wrappers may leak
only in static data (e.g. fields that were not cleared on deinit), but
we seem to have a race that breaks this assumption.
1. We do clean `all_lsp_adapters` field after
https://github.com/zed-industries/zed/pull/34334 and it's called for
every extension that is unregistered.
2. `LspStore::maintain_workspace_config` ->
`LspStore::refresh_workspace_configurations` chain is triggered
independently, apparently on `ToolchainStoreEvent::ToolchainActivated`
event which means somewhere behind there's potentially a Python code
that gets executed to activate the toolchian, making
`refresh_workspace_configurations` start timings unpredictable.
3. Seems that toolchain activation overlaps with plugin reload, as
`2025-07-28T12:16:19+03:00 INFO [extension_host] extensions updated.
loading 0, reloading 1, unloading 0` suggests in the issue logs.
The plugin reload seem to happen faster than workspace configuration
refresh in
c65da547c9/crates/project/src/lsp_store.rs (L7426-L7456)
as the language servers are just starting and take extra time to respond
to the notification.
At least one of the `.clone()`d `adapter`s there is the adapter that got
removed during plugin reload and has its channel closed, which causes a
panic later.
----------------------------
A good fix would be to re-architect the workspace refresh approach, same
as other accesses to the language server collections.
One way could be to use `Weak`-based structures instead, as definitely
the extension server data belongs to extension, not the `LspStore`.
This is quite a large undertaking near the extension core though, so is
not done yet.
Currently, to stop the excessive panics, no more `.expect` is done on
the channel result, as indeed, it now can be closed very dynamically.
This will result in more errors (and backtraces, presumably) printed in
the logs and no panics.
More logging and comments are added, and workspace querying is replaced
to the concurrent one: no need to wait until a previous server had
processed the notification to send the same to the next one.
Release Notes:
- Fixed warm-related panic happening during startup
This fixes an issue where we were not setting the context server working
directory at all.
Release Notes:
- Context servers will now be spawned in the currently active project
root.
---------
Co-authored-by: Danilo Leal <danilo@zed.dev>
Temporarily fixes#29133
Co-authored-by: Cole <cole@zed.dev>
Release Notes:
- python: Zed now reports a slightly different set of workspace folders
for Python projects to work around quirks in handling of multi-lsp
projects with virtual environment. This behavior will be revisited in a
near future.
Co-authored-by: Cole <cole@zed.dev>
In this PR we've reworked how git status updates are processed. Most
notable change is moving the processing into a background thread (and
splitting it across multiple background workers). We believe it is safe
to do so, as worktree events are not deterministic (fs updates are not
guaranteed to come in any order etc), so I've figured that git store
should not be overly order-reliant anyways.
Note that this PR does not solve perf issues wholesale - other parts of
the system are still slow to process stuff (which I plan to nuke soon).
Related to #34302
Release Notes:
- Improved Zed's performance in projects with large # of repositories
---------
Co-authored-by: Anthony Eid <hello@anthonyeid.me>
we were join(",") and split(",") to serialize the patterns.
This doesn't work when pattern includes a ","
example: *.{ts,tsx} (very common pattern used by agent)
help needed:
how will this work on version mismatch?
Release Notes:
- Fixed search filter patterns on remote projects.
This gets rid of the need to pass context to all cursor functions. In
practice context is always immutable when interacting with cursors.
A nicety of this is in the follow-up PR we will be able to implement
Iterator for all Cursors/filter cursors (hell, we may be able to get rid
of filter cursor altogether, as it is just a custom `filter` impl on
iterator trait).
Release Notes:
- N/A
Closes#31917
Previously, as of #28457 we used a hack, creating an empty transaction
in the history that we then merged formatting changes into in order to
correctly identify concurrent edits to the buffer while formatting was
happening. This caused issues with noop formatting however, as using the
normal API of the buffer history (in an albeit weird way) resulted in
the redo stack being cleared, regardless of whether the formatting
transaction included edits or not, which is the correct behavior in all
other contexts.
This PR fixes the redo issue by codifying the behavior formatting wants,
that being the ability to push an empty transaction to the history with
no other side-effects (i.e. clearing the redo stack) to detect
concurrent edits, with the tradeoff being that it must then manually
remove the transaction later if no changes occurred from the formatting.
The redo stack is still cleared when there are formatting edits, as the
individual format steps use the normal `{start,end}_transaction` methods
which clear the redo stack if the finished transaction isn't empty.
Release Notes:
- Fixed an issue where redo would not work after buffer formatting
(including formatting on save) when the formatting did not result in any
changes
Supersedes: #34242
Creates an `ActionArgumentsEditor` that implements the required logic to
have a JSON language server run when editing keybinds so that there is
auto-complete for action arguments.
This is the first time action argument schemas are required by
themselves rather than inlined in the keymap schema. Rather than add all
action schemas to the configuration options we send to the JSON LSP on
startup, this PR implements support for the
`vscode-json-language-server` extension to the LSP whereby the server
will request the client (Zed) to resolve URLs with URI schemes it does
not recognize, in our case `zed://`. This limits the impact on the size
of the configuration options to ~1KB as we send URLs for the language
server to resolve on demand rather than the schema itself. My
understanding is that this is how VSCode handles JSON schemas as well. I
plan to investigate converting the rest of our schema generation logic
to this method in a follow up PR.
Co-Authored-By: Cole <cole@zed.dev>
Release Notes:
- N/A *or* Added/Fixed/Improved ...
Closes#33510https://github.com/zed-extensions/php/issues/29
If certain language servers do not provide an insert/replace range, we
use `surrounding_word` as a fallback for that range, which internally
uses `word_characters`. It makes sense to use
`completion_query_characters` instead of `word_characters` to get that
range, because we use `completion_query_characters` to query completions
in the first place.
That means, for some hypothetical reason (e.g., if the Tailwind server
stops providing insert/replace ranges), we would correctly fall back to
the range "bg-blue-200^" instead of "200^", because
`completion_query_characters` includes "-" in this case.
For this particular fix, right now the default PHP language server
`phpactor` does not provide an insert/replace range, and hence
completion query character is used, which is `$` in this case.
Note that `$` isn't in word characters for reasons mentioned here:
https://github.com/zed-extensions/php/issues/14
Release Notes:
- Fixed an issue where accepting variable completion in PHP would result
in a double $ sign in the prefix.
when `terminal.detect_venv.activate_script` setting is default, pick the
appropriate activate script as per the `terminal.shell` settings
specified by the user. Previously when the activate_script setting is
default, zed always try to use the `activate` script, which only works
when the user shell is `bash or zsh`. But what if the user is using
`fish` shell in zed?
Release Notes:
- python: value of `activate_script` setting is now automatically
inferred based on the kind of shell the user is running with.
---------
Co-authored-by: Piotr Osiewicz <24362066+osiewicz@users.noreply.github.com>
Closes https://github.com/zed-industries/zed/issues/33980
Closes https://github.com/zed-industries/zed/discussions/33979
- Switches to the debounce task pattern for diagnostic summary
computations, which most importantly lets us do them only once when a
large number of DiagnosticUpdated events are received at once.
- Makes workspace diagnostic requests not time out if a partial result
is received.
- Makes diagnostics from workspace diagnostic partial results get
merged.
There might be some related areas where we're not fully complying with
the LSP spec but they may be outside the scope of what this PR should
include.
Release Notes:
- Added support for streaming LSP workspace diagnostics.
- Fixed editor freeze from large LSP workspace diagnostic responses.
Use `checked_sub` instead of checking for bounds manually. Also greatly
simplifies the logic for `next` and `previous`. Removing other manual
bounds checks as well
Release Notes:
- N/A
This PR adds a `severity` argument so severity can be defined when
navigating through diagnostics. This allows keybinds like the following:
```json
{
"] e": ["editor::GoToDiagnostic", { "severity": "error" }],
"[ e": ["editor::GoToDiagnostic", { "severity": "error" }]
}
```
I've added test comments and a test. Let me know if there's anything
else you need!
Release Notes:
- Add `severity` argument to `editor::GoToDiagnostic`,
`editor::GoToPreviousDiagnostic`, `project_panel::SelectNextDiagnostic`
and `project_panel::SelectPrevDiagnostic` actions
This is mostly setting up the UI for now; I expect it to be the biggest
chunk of work.
Release Notes:
- debugger: Added memory view
---------
Co-authored-by: Anthony Eid <hello@anthonyeid.me>
Co-authored-by: Mikayla Maki <mikayla.c.maki@gmail.com>
Co-authored-by: Mikayla Maki <mikayla@zed.dev>
Closes#33442
Release Notes:
- Resolved an issue where the ESLint language server returned an empty
string for the CodeDescription.href field in diagnostics, leading to
missing diagnostics in editor.
This solves problems where users couldn't shut down sessions while
locators or build tasks are running.
I renamed `debugger::Session::Mode` enum to `SessionState` to be more
clear when it's referenced in other crates. I also embedded the boot
task that is created in `SessionState::Building` variant. This allows
sessions to shut down all created threads in their boot process in a
clean and idiomatic way.
Finally, I added a method on terminal that allows killing the active
task.
Release Notes:
- Debugger: Allow shutting down debug sessions while they're booting up
This PR updates the debugger panel's session list to be more useful in
some cases that are commonly hit when using the JavaScript adapter. We
make two adjustments, which only apply to JavaScript sessions:
- For a child session that's the only child of a root session, we
collapse it with its parent. This imitates what VS Code does in the
"call stack" view for JavaScript sessions.
- When a session has exactly one thread, we label the session with that
thread's name, instead of the session label provided by the DAP. VS Code
also makes this adjustment, which surfaces more useful information when
working with browser sessions.
Closes#33072
Release Notes:
- debugger: Improved the appearance of JavaScript sessions in the debug
panel's session list.
---------
Co-authored-by: Julia <julia@zed.dev>
Co-authored-by: Remco Smits <djsmits12@gmail.com>
Fixes an issue that caused Windows to fail when removing extension's
directories, as Zed had never stop any related processes.
Now:
* Zed shuts down and waits until the end when the language servers are
shut down
* Adds `impl Drop for WasmExtension` where does
`self.tx.close_channel();` to stop a receiver loop that holds the "lock"
on the extension's work dir.
The extension was dropped, but the channel was not closed for some
reason.
* Does more unregistration to ensure `Arc<WasmExtension>` with the `tx`
does not leak further
* Tidies up the related errors which had never reported a problematic
path before
Release Notes:
- N/A
---------
Co-authored-by: Smit Barmase <heysmitbarmase@gmail.com>
Co-authored-by: Smit <smit@zed.dev>
Restarting sessions was broken in #33273 when we moved away from calling
`kill` in the shutdown sequence. This PR re-adds that `kill` call so
that old debug adapter processes will be cleaned up when sessions are
restarted within Zed. This doesn't re-introduce the issue that motivated
the original changes to the shutdown sequence, because we still send
Disconnect/Terminate to debug adapters when quitting Zed without killing
the process directly.
We also now remove manually-restarted sessions eagerly from the session
list.
Closes#33916
Release Notes:
- debugger: Fixed not being able to restart sessions for Debugpy and
other adapters that communicate over TCP.
- debugger: Fixed debug adapter processes not being cleaned up.
---------
Co-authored-by: Remco Smits <djsmits12@gmail.com>
Previously if I set enabled: false for one the context servers in
settings.json it will not show up in the settings in agent panel when I
start zed. But if I enabled it from settings it properly showed up. We
were filtering the configuration to only get the enabled context servers
from settings.json. This PR adds fetching all of them.
Release Notes:
- agent: Show context servers which are disabled in settings in agent
panel settings.
